Application of Neural Networks to Image Encryption

Abstract

One of the major aspects of computer systems is cryptography. In this work, we are going to take general ideas of the basics of cryptography systems. And since emerging technologies are not complete without neural networks, and current research has demonstrated that these systems can be used for a wide range of applications, we will talk about the basics of neural networks in cryptography and their domain of application. In this work, we introduce a system of Six-Dimensional Cellular Neural Network (6D-CNN) to generate a pseudo-random number. With additional proposed improvement method to generate the initial key conditions of the 6D-CNN using Lorenz 3D and Chen’s 3D systems. The presented 6D-CNN has hyper-chaos characteristics, very good sensitivity to initial conditions, and excellent randomness. We proposed a novel image encryption algorithm based on chaos and CNN, where it takes the architecture of chaos based on a substitution-diffusion image encryption cryptosystem. In the substitution part, we shuffle the image pixels coordinates, whereas in diffusion we use the generated pseudo-random sequence by 6D-CNN to encrypt the shuffled image. Finally, we run a security evaluation for our cryptosystem to assure its ability to offer the necessary security. Additionally, we compared our scheme to other related works to express its efficiency.